Anyone wanting a peaceful sleep should avoid the latest annual report from the Bank for International Settlements.
BIS has a decent record of spotting trouble before everyone else gets the memo, and the report, released on 28 June 2026, spells out how an AI bust could shove the global financial system into disorder.
This time, the pop could start with tech outfits pouring too much cash into AI infrastructure. Once investors notice the returns are not matching the vast spending, financing could dry up and share prices could take a kicking.
None of this is especially fresh, but the BIS joins it to a nastier picture of consumers, governments and the global economy.
US households are far more exposed to stocks than they were over the past two decades, relative to both wealth and income. Rising US share prices during the AI boom have given US-listed companies a huge weight in global indexes.
That means a US tech bust could chew through wealth well beyond Silicon Valley’s glassy temples. The BIS notes that tech companies are only the visible tip of a much larger supply chain iceberg.
If they slow capital spending, construction contractors with weaker balance sheets could feel the pain quickly. Other financial dominoes are already standing around, looking nervous and badly insured.
The debt increasingly fuelling AI investments could worsen the damage. Advanced economy governments are already stretched, leaving them with less room to revive activity through spending.
Persistently high inflation and geopolitical uncertainty stemming from the war in Iran make life harder for central banks seeking a clean response, the report said.
There is a fair chance the BIS is right again about AI, which has looked like a spending and chip-revenue bubble for a while. Exactly when it pops and which corporate genius is holding the bag are harder to guess.
Meanwhile, South Korea’s memory chip giants Samsung Electronics and SK Hynix are behaving as if there is no AI bubble. The companies are planning a $520 billion memory factory hub in the country’s south-west.
That would massively expand production capacity during a period of daft AI-driven profits. The planned plants will not start production for several years.
By then, Samsung and SK Hynix must hope AI customers still want endless memory chips rather than cheaper excuses.
